Small-Scale Farmers’ Preference Heterogeneity for Green Agriculture Policy Incentives Identified by Choice Experiment

Abstract

This study addresses differentiation among small-scale farmers’ preferences for green agriculture policy incentive mixes. Transforming modern agriculture to ecological fertilization and pest extermination practices is paramount in developing green agriculture, but policy incentives aimed at stimulating small-scale farmers’ adoption of ecological fertilization and deinsectization techniques are often challenged by those farmers’ heterogeneous characteristics and their consequent mixed incentive preferences. We establish a model examining the interplay between small-scale farmers’ characteristics (e.g., age, education level, family size, participation in agricultural organization) and combinations of incentive policies (i.e., green subsidy, technical support, environmental propaganda, agricultural insurance) in farmers’ willingness to participate in ecological fertilization/deinsectization, using a sample of 1032 Chinese farmers. By applying a mixed logit model and latent class model regressions, we find that farmers’ age, education level, family size, and farming organization participation are the most important characteristics influencing farmers’ preferences. Specifically, senior farmers tend to accept an incentive policy combination of green subsidy and technical support; farmers with higher education levels prefer an incentive policy combination of technical support and environmental propaganda; and larger families prefer an incentive policy combination of technical support and agricultural insurance. Additionally, participation in any agricultural organization reduces the household’s preference for incentive policy combinations of technical support, agricultural insurance, and green subsidy. Based on these findings, a typology of small farmers’ green agriculture incentive preferences (including security, monetary, and autonomy orientations) is proposed, offering suggestions for future green agriculture policy optimization.

1. Introduction

The practice of green agriculture, which has important implications for the conservation of natural resources [1,2] and the circulation of green end-products [3,4], is a critical component of the overall sustainability movement [5,6,7]. The accumulative alternation of chemical fertilizers and pesticides via ecological techniques is a pivotal agenda of the green agriculture process [6]. However, farmers have been historically motivated to adopt the most efficient fertilization and deinsectization techniques (i.e., chemical methods), which potentially have negative effects on the external environment, including land and water pollution as a result of the overuse of chemical fertilizers and pesticides [8]. Therefore, changing farmers’ payoffs and intervening in their fertilization and pest-eradication decisions through various policy incentives appears necessary during such a green agriculture process [9]. Nevertheless, despite the attention that the stimulation and adoption of green agriculture techniques has elicited from researchers and practitioners in recent times (see, e.g., [10,11,12,13]), several challenges, such as inaccessibility to technical/financial support and concerns about economic risks have emerged, and the stimulation of green agriculture techniques will not necessarily lead to their adoption, especially among small-scale farmers [14,15]. However, extant research and practice usually focus on optimizing a specific policy incentive (e.g., green subsidy; see [16,17]), rather than an interplaying incentive set, and treat small-scale farmers as characteristically homogenous objects rather than subjects with heterogeneous characteristics. Considering the reduced emphasis on small-scale farmers’ characteristic heterogeneity, leading to insufficient discussion of preference heterogeneity of green agriculture incentives in extant literature, the current study, through a sample of 1032 small-scale farming households in China, addresses the influence of heterogeneity in farmers’ characteristics on their preference for policy incentive sets consisting of different arrangements of incentive attributes.

2. Research Background

2.1. Green Agriculture and Ecological Fertilization and Deinsectization

In 2015, the United Nations Development Programme launched 17 Sustainable Development Goals as a universal call to action to achieve balanced and sustainable development socially, economically, and environmentally [18]. Green agriculture is a crucial part of this global sustainable development blueprint [19]. Many nations have made reasonable attempts to realize green agriculture, including ecologically friendly farmland circulation emphasizing the maintenance of farmland quantities (see, e.g., the Conservation Reserve Program of the United States [20]; Common Agricultural Policy in the European Union [21]; Summer Fallow in Australia [22]; the Ecological Farming System of Farmland Circulation in China [23]) and measures aimed at maintaining the farmland quality, especially avoiding farmland pollution. There are numerous pollution sources in farmland, such as chemical fertilizers and pesticides [24,25], agriculture film [26], municipal refuse [27], and industrial sewage [28]. Chemical fertilizers and pesticides are primary pollution sources [25,29,30], which many nations are attempting to reduce through their green agriculture processes.

Fertilization and deinsectization are essential farming routines highly correlated with both farming outputs (i.e., crop yield, farmers’ income) and farming externalities (i.e., soil maintenance and farmland pollution) [31]. In many developing countries, fertilization and pest eradication are still dominated by the use of chemical fertilizers and pesticides [32] because of their effectiveness in increasing crop yields and, thus, farmers’ income. However, chemical fertilizers and pesticides are widely acknowledged as causing severe environmental problems such as soil degradation and intensified agricultural non-point source pollution [30,32]. Thus, transforming farmers’ conventional adoption of chemical fertilizers and pesticides to ecological fertilization and deinsectization techniques has become crucial in agriculture development in many developing countries [33,34].

However, promoting ecological fertilization and pest eradication has historically been challenged by farmers’ (especially small-scale farmers’) insensitivity to incentive measures (see, e.g., [35])—for instance, in China, where the small-scale farming household is the historical basic unit of China’s agricultural production. Because the cultivated land resource per unit population is very limited in this heavily populated nation, Chinese farmers are habituated to intensive utilization of chemical fertilizers and pesticides in order to increase agricultural productivity. It is noteworthy that the economic considerations that dictate farmers’ choices generate natural externalities for the environment. Compared with other production and operation entities, small-scale farmers generally have a low level of education, with little knowledge of pesticide application and relatively low awareness of production safety and subjective sense of responsibility, resulting in improper fertilizer and pesticide application practices [34,36]. Ecological farming (at least in the extant stage) restricts the land property rights of farmers carrying out ecological agriculture and affects their land income [37,38]. Therefore, it is undeniably crucial for the government to implement policy incentives promoting the uptake of alternative practices for farmers conducting ecological fertilization and pest eradication [26,38].

2.2. Policy Incentives for Ecological Fertilization and Deinsectization

Policies provide important guidance and guarantees for farming households’ behavior. In the previous literature, the most commonly used policy measures to guide farmers’ scientific or reduced application of chemical fertilizers and pesticides have been divided into two categories: economic incentives, such as biopesticide subsidies [39] and agricultural insurance [40,41]; and voluntary incentives, including technical support and environmental propaganda [34].

Green subsidies promote the application of biofertilizers and biopesticides through the provision of allowances to farmers [39]. Some scholars have found that subsidy policies can effectively motivate farmers to adopt green agriculture technologies (see, e.g., renewable soil fertility replenishment technologies [42]). Whether the government subsidizes production is an important consideration for farmers in their decision making with regard to the adoption of biopesticide application [43,44]. However, some scholars argue that subsidies function effectively only alongside other incentive policies [38,42,45]. Agricultural insurance policies involve insurance (provided by insurance institutions) to compensate for property losses caused by natural disasters, accidents, epidemics, diseases, and other accidents covered by the agricultural insurance contract, thus helping farmers cope with multiple risks inherent to farming activities and to compensate their income loss under uncertainty [41,46]. Technical support refers to the provision of technical training, consultation with farmers, and promoting green agriculture technology innovation among them [34]. Finally, environmental propaganda refers to disseminating knowledge of the concepts of sustainable development, ecological variety, environmental protection, and resource conservation to farmers [34]. Studies have shown that technical support [47], environmental propaganda [48], and agricultural insurance [49] have a prominent effect on the uptake of green agriculture technologies such as replacing the use of chemicals with ecological methods. However, some other studies indicate a suspicion that these incentives may be not significant [34,50].

2.3. Farmers’ Choices of Fertilization and Deinsectization Techniques

Examining farmers’ adoption of ecological fertilizers and biopesticides, the previous literature identifies the following influencing factors.

Individual-level characteristics. Among the various characteristics, age and education level [51,52] are the most well researched. For instance, it has been shown that senior farmers are less enthusiastic towards adopting ecological farming techniques than the new generation of farmers [51,52]. More educated farmers tend to adopt ecological farming techniques more easily [53]. In addition, recent studies have begun to focus on the influences of factors such as gender [52], risk preference [54], years of farming [55], and social capital [56].

Household-level characteristics. Previous research has also indicated that the family’s resource endowment (e.g., family population, family income, farmland scale, participation in farming organizations, livelihood mode, and family assets) could influence the choice to adopt ecological farming. For instance, total household income plays a positive role in small farmers’ choice to adopt green agriculture [52,57]. The distribution of the labor force also affects their preferences [52]. The rise of farmland scale causes a deduction in chemical fertilizer usage [52,58]. In addition, whether the household joins in farming organizations, or its livelihood mode differentiation [59,60], significantly impact environment-friendly technology acquisition behavior.

Environmental characteristics. In addition to these main influencing factors, the natural (e.g., farmland location [61]; climate and soil condition [62]; planting diversity [63]; farmland distance from the marketplace [64] or from the highway [55]; farmers’ indigenous traditional knowledge of their relations with the living environment [65,66]), social (neighborhood communication [67]; social network [68]), and economic environment (existence of quality standards and pesticide residue detection [69]; availability of agricultural technology extension services and loans [70]; the convenience of purchasing chemical fertilizers and pesticides [71]; stakeholder influences such as influence from academic institutions, non-profit organizations, or the buyer [72]) can also have an impact on farmers’ green agriculture behavior choice.

2.4. Research Gap

Despite extant research investigating the effects of green agriculture incentive policies, the majority of studies only validate or illustrate a specific policy effect (see, e.g., [43,44,47,48,49]). Synthesis and inter-comparison of these policies is lacking, fragmenting the effects of their interplay. Moreover, extant research invalidating and illustrating policy effects from a macroscopic perspective (based on regional statistical data) is limited; in addition, policies’ impact on the changes in small-scale farmers’ preferences, from a microscopic perspective, is scarce—in particular, considering that heterogeneity in farmers’ characteristics will moderate the policy effects in the incentive-response model. However, previous research usually treats farmers as homogenous objects without distinct characteristics rather than objects with a heterogeneity of characteristics. Although farmers’ heterogeneity of characteristics is beginning to be studied, only a few characteristics have been examined, rather than a collation of multiple characteristics or characteristics of different hierarchies. In the current study, we argue that the conventional incentive-response model investigating policy incentives’ effects on farmers’ green agriculture adoption should be complemented by incorporating farmers’ heterogeneity of characteristics at both the individual and household level.

3. Methodology

3.1. Overall Research Design

In this paper, we performed a choice experiment [73,74] to collect data on characteristics and preference from individual farmers. We then evaluated the relationship between farmers’ choices to abandon chemical fertilizers/pesticides, their characteristics, and the policy incentive attribute combination they face, via regression evaluation based on the mixed logit model (MLM) [75] and latent class model (LCM) [76] (refer to Figure 1). Specifically, according to Lancaster’s stochastic utility theory [77], farmers’ policy incentive mix preference directly affects implementation effect; conversely, the optimal policy incentive mix is the one that farmers prefer and choose. Therefore, in this study, farmers were asked to select from multiple scenarios with different attributes (choice experiment), and their preference for these attributes was determined through the regression estimations. The MLM reveals the heterogeneous preferences of farmers for incentive policies on chemical reduction, while the LCM classifies farmers with heterogeneous preferences into several categories.

3.2. Variables Setting

Small-scale farmers’ willingness to participate (WTP), or their preference for the specific incentive policy mix was set as the outcome variable in this study. The measurement of WTP is further specified in Section 3.3. Dependent variables are composed of farmers’ characteristics (i.e., the householder’s gender, age, education level, family size, family’s agricultural laborers, total planting area, grain-planting motivation, grain-based income proportion, and participation in agricultural organizations—according to a summary of previously researched farmers’ characteristics) and the aforementioned four incentive policies (i.e., biopesticide subsidies, agricultural insurance, technical support, and environmental propaganda). Additionally, chemical abandon rate, referring to the rate of abandonment of the usage of chemical fertilizers and pesticides, is also included in the research model to introduce the scale effect of green agriculture technique adoption. The measurement of dependent variables is summarized in

Table 1. Measurements of dependent variables.

Categories	Subcategories	Variables	Abbreviation	Levels/Measures
Farmers’ characteristics	Individual-level characteristics	Age (years)	AGE	<35; 35–45; 46–55; 55–65; >65
		Education level	EDUC	Illiterate; Primary school; Middle school; High school; College or above
	Household-level characteristics	Family size (person)	FAMI	Measured by person counts
		Agricultural labor (person)	LABO	Measured by person counts
		Planting area (hm2)	PLAN	Measured by hectometer square
		Grain-planting motivation	MOTI	Self-supply; Self-supply & income increase; Income increase
		Grain-based income proportion (%)	INCO	Measured by the proportion of grain-based income in the overall household income
	Environmental characteristics	Participation in agricultural organizations	ORGA	0: does not participate in any agricultural organizations 1: participates in some agricultural organization
Green agriculture incentives	Economic incentives	Green subsidy	SUBS	0: without biopesticide subsidy 1: with a medium level of subsidy such as CNY 5/mu 2: with a high level of subsidy such as CNY 10/mu
		Agricultural insurance	INSU	0: without agricultural insurance 1: with agricultural insurance
	Voluntary incentives	Technical support	TECH	0: without technical support 1: with technical support
		Environmental propaganda	ENVI	0: without environmental propaganda 1: with environmental propaganda
Scale effect indicator	—	Scale of change	SCAL	0: not changing chemical fertilizers and pesticides usage 1: changing chemical fertilizers and pesticides usage at a medium level of 5% 2: changing chemical fertilizers and pesticides usage at a high level of 15%

.

3.3. Theoretical Analysis Model

3.3.1. Lancaster Stochastic Utility Model

The choice experiment is established based on Lancaster’s stochastic utility theory. As assumed in this theory, farmer n can choose solution i from choice set $C_n$ of a specific policy attribute, and the indirect utility $U_{ni}$ of this choice is greater than that of other alternatives. Thus, the indirect utility function can be expressed as:

$$U_{ni}=V_{ni}\left(Z_{ni},S_{ni}\right)+{\epsilon }_{ni}$$

(1)

The indirect utility function $U_{ni}$ can be divided into the observable utility $V_{ni}$ and stochastic utility ${\epsilon }_{ni}$ (representing the utility from unobservable factors). The observable utility $V_{ni}$ can be further expressed as a function with the policy attribute and the farmer characteristic variables $S_{ni}$.

The policy attribute variables, policy outcome variables, farmer characteristic variables, and interaction items can be included in the formula of observable utility $V_{ni}$:

$$V_{ni}=ASC+{\gamma }_{n}Outcom{e}_{ni}+{\sum }_k{\beta }_{ik}{x}_{ik}+{\sum }_k{\lambda }_k{x}_{nik}{x}_{nik}+{\sum }_{k,h}{\alpha }_h{S}_{nh}{x}_{nik}+{\sum }_{h}AS{C}_n{S}_{nh}$$

(2)

In Equation (2), ASC is a substitute for a specific constant term: if farmers select any of the policy incentives, the value of ASC is assigned as 1; if not, the value is assigned as 0. ${\gamma }_n$ is the coefficient of $Outcom{e}_{ni}$. ${\beta }_{ik}$ is the coefficient of policy attribute set $x_{ik}$. $S_{nh}$ is the h_th basic characteristic variable of the farmer n. $x_{nik}{x}_{nik}$, $S_{nh}{x}_{nik}$, and $AS{C}_n{S}_{nh}$ represent the interaction items among policy attributes, between farmer characteristic attributes and policy attributes, and between constant term ASC and farmer characteristic attributes, respectively. ${\gamma }_k$ is the coefficient of interaction items among policy attributes. ${\alpha }_h$ is the coefficient between policy attribute and characteristic variables.

3.3.2. Farmers’ Preference Heterogeneity

This study further defines a parameter ${\beta }_n$ referring to the coefficient of a specific incentive mix, or, when farmer n is in a specific choice set $C_n$. $f\left({\beta }_n\right)$ is its density function and ${\beta }_n$ obeys the stochastic distribution. Considering that farmers’ preferences were strongly heterogeneous, this study applied MLM to loosen the assumption that variables have identical independent distribution; therefore, the attribute parameters were enabled to change randomly among different farmers. Specifically, the probability of farmer $\mathrm{n}$ choosing the $i_{\mathrm{th}}$ policy attribute set can be expressed by the MLM as follows:

$$P_{ni}=\int \frac{\exp \left(\beta X_{ni}\right)}{{\sum }_j\exp \left(\beta X_{ni}\right)}f(\beta |\theta )d\beta$$

(3)

In Equation (3), the probability density $\beta$ can be expressed as a random variable following the distribution $f(\beta |\theta )$, and 0 is the real parameter describing the distribution.

If the distribution $f(\beta |\theta )$ is discrete, Equation (3) can be further transformed into an LCM to classify N farmers into S different categories by their preference similarity. Specifically, we construct $R_{ns}$, the probability of farmer $\mathrm{n}$ falling into the sth latent class, as:

$$R_{ns}=\frac{\exp \left({\mu }_s{z}_n\right)}{{\sum }_s\exp \left({\mu }_s{z}_n\right)}$$

(4)

In Equation (4), ${\mu }_s$ is the parameter vector of farmers in the $s_{\mathrm{th}}$ latent class and $z_n$ is a series of feature vectors that influence the farmer $n$ to fall into a latent class. Therefore, the probability of farmer n falling into the ith latent class and choosing the ${\mathrm{i}}_{\mathrm{th}}$ policy attribute set can be further expressed as:

$$P_{ni}={\sum }_s\frac{\exp \left(\beta X_{ni}\right)}{{\sum }_j\exp \left(\beta X_{ni}\right)}{R}_{ns}$$

(5)

In which ${\beta }_s$ is the parameter vector of farmers in the sth category. Therefore, the unconditional probability of solution i can be expressed as:

$$P(i|C_n)=\int \frac{e^{V_{ni}}}{{\sum }_j{e}^{V_{ni}}}f\left({\beta }_n\right)d{\beta }_n,\forall j\in C_n$$

(6)

3.3.3. Constructing Willingness to Participate

According to Lancaster’s stochastic utility theory, farmers’ WTP can be expressed as the chemical abandon rate that farmers are willing to accept to maintain the same utility when the attribute of an incentive policy changes, namely, the marginal substitution rate.

$$WT{P}_k=\frac{\partial V_{ni}∕\partial {\gamma }_n}{\partial V_{ni}∕\partial {\beta }_k}$$

(7)

In Equation (7), ${\gamma }_n$ is the variable coefficient of the policy outcome, and ${\beta }_k$ is the variable coefficient of the policy attribute. The larger the $WT{P}_k$, the greater is the chemical abandon rate that farmers are willing to accept under the incentive policy, which reflects the effect of the incentive policy.

3.4. Choice Experiment Process

This paper uses a choice experiment to study the influence of ecological farming rules on farmers’ choice of ecological farming intention. The choice experiment method expresses individual behavior in terms of discrete choice under the framework of utility maximization [78]. It provides individuals a selection of different properties, lets them choose the best scheme from each choice, and then analyzes the value of the different attributes and various solutions of different properties of relative value [78]. The experimental method avoids the embedded and strategic biases caused by the contingent valuation method [78].

Based on the full factorial design for the aforementioned five policy attributes and their corresponding attribute levels (as listed in Table 1), 2 × 2 × 2 × 3 × 3 = 72 options can be obtained, generating ${\mathrm{C}}_{72}^2$ = 2556 choice experiment tasks through pairwise combination. It is unrealistic for the implementer to make a comparative choice between over two thousand experimental tasks. Generally speaking, the implementer will become tired when they identify approximately 15–20 choice experiment tasks, and their selection efficiency and validity will seriously deteriorate [79]. Thus, this study employs the fractional factorial design as an appropriate solution [79].

The minimum number of tasks (n = 27) was obtained by using the OPTEX program in Statistics Analysis System (SAS, version 9.4, SAS Institute Inc., Cary, NC, USA), together with the Choice-Based Conjoint Analysis (CBC, version 1.0.1, Sawtooth Software Inc., Provo, UT, USA). The 27 choice experiment tasks were further randomly divided into nine groups, and nine different versions of the choice experiment questionnaire were generated. Each questionnaire version contains three different choice experiment tasks, and each task includes two different solutions plus one null choice (i.e., neither of the two solutions is selected) (refer to

Table A1. Sample of Choice Experiment Tasks.

Attributes	Solution A	Solution B	Solution C
Technical Support	None	None	Neither of the previous two solutions are selected
Environmental Propaganda	Yes	None
Agricultural insurance	Yes	None
Green subsidy	None	CNY 10/mu
Change Rate of Chemical Usage	Down by 5%	Down by 15%
Your Choice (tick “√”)

for the arrangement of one experiment task sample). Implementers were asked to choose their preferred policy mix from each experimental task.

3.5. Research Sample

In this study, farming households in Hubei Province, China, were selected as the research sample for the following reasons. First, green agriculture is a major direction for Chinese agriculture development and the solution for the problems facing China, such as deteriorating land quality and land yields. The Chinese government has focused on maintaining the quantity of farmland (e.g., through agriculture land circulation programs [23]). It is not until recent years that the Chinese government has begun to emphasize the significance of maintaining farmland quality in the face of significant soil fertility decline, pollution, and potential risks to food security and sustainability development, and has thus devoted attention to exploring policies for farmland quality maintenance. For instance, the Central Committee of China’s Communist Party has released Advice for the 13th Five-Year Plan for National Economic and Social Development, pointing out that due to the country’s long-term development, China has been facing problems including excessively high land-use intensity, serious decline in soil fertility, soil and water loss, extensive exploitation of groundwater resources, soil degradation, and non-point source pollution, all of which have become challenges restricting sustainable agricultural development. Second, a major reason for those problems is the overwhelming use of chemical products, such as chemical fertilizers and pesticides. The amount of pesticide applied in China increased from 765,000 tons (in 2009) to 1,754,000 tons (in 2019), at a growth rate of 129.8%. The amount of pesticide applied per unit area also increased year by year, from 7.66 kg/hm² in 1990 to 13.81 kg/hm² in 2008, twice that in developed countries. In terms of chemical fertilizer, the amount of fertilizer applied nationwide increased from 8.84 million tons in 1988 to 60.5 million tons in 2019, an increase of nearly seven times; the amount of fertilizer applied per unit area in China was 379.5 kg/hm² in 2019, far higher than the global average amount of fertilizer 120 kg/hm², and also exceeding the international safe fertilization limit of 225 kg/hm². Additionally, as one of the 13 major grain-producing provinces in China, Hubei Province has rich cultivated land and grain resources, convenient transportation, a good foundation for economic and social development, and good cultivated land resource protection.

Questionnaires were distributed to 1100 grain-planting farmers (also householders) in the cities of Huanggang and Xiangyang in Hubei Province from March to April 2021 (refer to Figure 2). One thousand and thirty-two valid answers were retained after excluding incomplete and carelessly completed answers (response rate = 93.81%). Basic information for the respondents is shown in

Table 2. Basic statistical characteristics of sample farmers.

Type	Option	Sample Size	Percentage (%)
Age (years)	<35	38	5.09
	35~45	108	14.48
	45~55	234	31.37
	55–65	233	31.23
	≥65	133	17.83
Education level	Illiterate	26	3.49
	Primary school	205	27.48
	Middle school	328	43.97
	High school	164	21.98
	College or above	23	3.08
Family size (person)	1–2	56	5.43
	3–4	467	45.25
	5–6	413	40.02
	7–8	61	5.91
	9–10	35	3.40
Agricultural labor (person)	0	24	3.22
	1–2	596	79.89
	3–4	118	15.82
	5–6	8	1.07
Planting area (hm2)	<0.67	237	31.77
	0.67–3.33	254	34.05
	3.33–6.67	110	14.75
	6.67–13.33	64	8.58
	≥13.33	81	10.86
Grain-planting motivation	Self-supply	97	13.00
	Self-supply & income increase	317	42.49
	Income increase	332	44.50
Grain-based income proportion (%)	<10%	147	19.71
	10–30%	133	17.83
	30–50%	127	17.02
	≥50	339	45.44
Participation in agricultural organizations	Yes	384	37.21
	No	648	62.79

: 62.6% of the total respondents were 45–65 years old; 71.45% of the total sample received primary and secondary education. Families with three or four members occupied the largest proportion (45.25%) of the sample, and 83.11% of the households had two or fewer than two agriculture laborers. The grain production scale of the households was relatively small, and 65.82% had a planting area smaller than 3.33 hm². Their motivation for growing grain was self-sufficiency (13.00%) and income increase (44.50%), or a mixture of the two (42.49%). For 54.56% of the sample, income earned from grain planting accounted for less than 50% of the total household income, indicating diversity in the rural households’ income. In addition, 62.79% of the respondents had never participated in any agriculture organizations.

4. Results

4.1. Estimation Results of Mixed Logit Model

This study used the Nlogit software (version 5.0, Econometric Software Inc., New York, NY, USA) to conduct an estimation of MLM, and the estimation results are shown in

Table 3. Estimation results of Mixed Logit Model.

Variables	Model 1		Model 2		Model 3
	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.
ASC	−0.419 **	0.168	−5.677 ***	2.202	−0.631 **	0.248
SUBS	0.115 ***	0.014	1.933 ***	0.614	0.011	0.066
INSU	0.304 ***	0.048	3.016 **	1.444	0.331***	0.125
TECH	0.279 ***	0.049	4.672 **	1.993	0.188	0.253
ENVI	0.101 ***	0.039	1.428	1.213	0.166 ***	0.054
SCAL	0.013 **	0.006	0.380 **	0.159	0.715	0.863
SUBS × INSU	—	—	0.561 **	0.250	—	—
SUBS × TECH	—	—	−0.016	0.216	—	—
SUBS × ENVI	—	—	0.113	0.191	—	—
INSU × TECH	—	—	0.432	0.773	—	—
INSU × ENVI	—	—	1.437 *	0.873	—	—
TECH × ENVI	—	—	−1.722 *	0.950	—	—
EDUC × ASC × TECH × ENVI	—	—	—	—	0.218 **	0.216
FAMI × ASC × INSU × TECH	—	—	—	—	0.112 **	0.152
AGE × SUBS × TECH	—	—	—	—	0.063 **	0.083
ORGA × SUBS × INSU × TECH	—	—	—	—	−0.687 ***	0.332
Log likelihood	−3295.837		−3091.117		−3089.734
$\mathrm{McFadden} \mathrm{Pseudo} R^2$	0.095		0.101		0.102

Notes: S.E. = Standard Error. *, ** and *** indicate significant results at the levels of 10%, 5% and 1%, respectively. Restricted by the length of the paper, in Model 3, only statistically significant values are listed as the interactive term coefficient between ASC and characteristic variables of farmers as well as the cross coefficient between attribute variables of the incentive policy and characteristic variables of farmers.

. Specifically, Model 1 contains only incentive policy attribute variables as dependent variables (without any interactions); Model 2 includes interaction items between two incentive policy attribute variables ahead of Model 1; and Model 3 further complements Model 2 by adding interaction items between farmer characteristic variables and policy incentive mix. All three models have been appropriately fitted (Pseudo-R² = 0.095). The coefficients of ASC are significantly negative in the three models (β = −0.419, S.E. = 0.168 in Model 1; β = −5.677, S.E. = 2.202 in Model 2; β = −0.631, S.E. = 0.248 in Model 3), indicating that the implementers generally have a tendency to deviate from the status quo; that is, farmers are generally willing to make changes (reduce the use of chemicals) according to the relevant incentive policies, in order to obtain a higher level of utility. Moreover, most of the incentive policy attribute variables of those four models have significant utility coefficients, indicating the existence of heterogeneity in farmers’ preference for incentive policies. The following subsection further illustrates how the incentive policy attributes influence the chemical abandon rate.

In Model 1, the coefficients of green subsidy (β = 0.115, S.E. = 0.014), agricultural insurance (β = 0.304, S.E. = 0.048), technical support (β = 0.279, S.E. = 0.049), and environmental propaganda (β = 0.101, S.E. = 0.039) are all significantly positive, indicating that these incentive policies can effectively improve the utility level of farmers’ policy mix preference and consequently elicit a significant change in chemical usage reduction. However, the coefficient of environmental propaganda (β = 1.428, S.E. = 1.213) is no longer significantly positive in Model 2, nor are those of green subsidy (β = 0.011, S.E. = 0.066) and technical support (β = 0.188, S.E. = 0.253) in Model 3, indicating that the interactions between those policy incentives alters their isolated effects (specified as follows).

As for the interaction between incentive policies, the interaction coefficients between green subsidy and agricultural insurance (β = 0.561, S.E. = 0.250 in Model 2) is significantly positive, indicating a complementary relationship between the two incentive policies. Similarly, a complementary relationship exists between environmental propaganda and agricultural insurance, because the interaction coefficient (β = 1.437, S.E. = 0.873 in Model 2) is significantly positive. However, the interaction coefficients between technical support and environmental propaganda (β = −1.722, S.E. = 0.950 in Model 2) are significantly negative, indicating a substitution relationship between the two incentive policies. None of the other interaction coefficients between incentive policy attributes are significant.

In terms of interplay between farmer characteristics and the incentive policy mix, Model 3 in Table 3 shows that the interaction coefficient between farmers’ education level, ASC, technical support, and environmental propaganda (β = 0.218, S.E. = 0.216) is significantly positive, indicating that farmers with higher education levels prefer the incentive policy combination of technical support and environmental propaganda. Similarly, the interaction coefficient between family size, ASC, agricultural insurance, and technical support (β = 0.112, S.E. = 0.152) is significantly positive, indicating that farmers with larger families prefer the incentive policy combination of agricultural insurance and technical support. The significantly positive interaction coefficient between age, green subsidy, and technical support (β = 0.003, S.E. = 0.001) implies that senior farmers tend to accept an incentive policy combination of green subsidy and technical support. However, the interaction coefficient between organization participation, green subsidy, agricultural insurance, and technical support (β = −0.687, S.E. = 0.332) is significantly negative, indicating that households that have participated in any form of agricultural organization do not accept the mix of those three incentives.

4.2. Estimation Results of Latent Class Model

The latent class model has been applied in this paper to analyze the preference heterogeneity of farmers in different groups. Akakchi Information Criteria (AIC) and Bayesian Information Criteria (BIC) are used to measure the fitting effect of the LCM in order to decide the most appropriate categorizing numbers. As shown in

Table 4. Multi-classification indexes of the Latent Class Model.

Index	Category-2	Category-3	Category-4
Log likelihood	−3100.911	−3053.659	−3042.494
Number	13.000	20.000	27.000
McFadden Pseudo $R^2$	0.099	0.112	0.115
AIC	6227.800	6147.300	6139.000
BIC	6261.200	6207.200	6228.000

, when the number of categories is three, the LCM has a minimum BIC value. Therefore, in this study, the category number was set as three. Estimation results of LCM are shown in

Table 5. Estimation results of the Latent Class Model.

Variables	Group 1		Group 2		Group 3
	Economy-Orientated		Security-Orientated		Autonomy-Orientated
	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.
ASC	−4.308 ***	1.295	−21.756 *	11.589	2.876 **	0.523
Green subsidy	0.623 ***	0.161	0.492 *	0.295	0.273 ***	0.049
Agriculture insurance	−3.089 ***	0.773	20.367 *	10.960	0.470 ***	0.163
Technology support	0.121 **	0.159	10.935	6.173	−0.395 **	0.165
Environmental propaganda	−1.951 ***	0.476	8.456 **	4.238	0.885 ***	0.185
Scale of change	−0.266 ***	0.065	1.31 *	0.733	0.056 **	0.022
Group proportion	38.5%		36.5%		25.0%
Characteristics	Mean	S.D.	Mean	S.D.	Mean	S.D.
Age (years)	47	26	62	12	58	17
Family size (person)	5	6	6	5	5	6
Agricultural labor (person)	4	4	5	4	3	5
Planting area (hm2)	4.67	3.17	5.63	2.36	6.31	1.89
Log likelihood	−3053.659		McFadden Pseudo $R^2$		0.112

Notes: S.E. = Standard Error. S.D. = Standard Deviation. *, ** and *** indicate significant results at the levels of 10%, 5% and 1%, respectively.

.

According to Table 5, farmers can be divided into three groups. For the first group, which holds 36.5% of the whole population, the coefficient of technical support is non-significant, while the coefficients of the other three are all positively significant: the largest being for agricultural insurance (β = 20.367, S.E. = 10.960), followed by environmental propaganda (β = 8.456, S.E. = 4.238), and then marginally by green subsidy (β = 0.492, S.E. = 0.295). In addition, the coefficient of chemical abandon rate is significantly positive (β = 130.916, S.E. = 73.257). We term this group as the economy-orientated group because the adoption motivation for green technology for these farmers seems to be mostly attributable to the affordability of bio-insurance. This economy-orientated group has a typical profile of a household head of younger age (M_age = 47) and higher-level education (high school), a modest family size of five people, yet with a medium agricultural labor population of four people, and a small planting area at 4.67 hm².

The second group (38.5% of the whole population) has significantly positive coefficients for green subsidy (β = 0.623, S.E. = 0.161) and technical support (β = 0.121, S.E. = 0.159), and significantly negative coefficients for chemical abandon rate (β = −26.568, S.E. = 6.462), agricultural insurance (β = −3.089, S.E. = 0.773), and environmental propaganda (β = −1.951, S.E. = 0.476). Considering the negative scale effect in this group, and that the most influenced incentive is green subsidy, this group is termed the security-orientated group. Such a group is typically characterized as the largest family (M_family = 6) with the most agriculture labor population (M_labor = 5), but a medium planting area (M_planting = 5.63 hm²), and an older (M_age = 62) and middle-school-educated household head.

The last type occupies 25.0% of the whole sample. Despite the coefficient of technical support (β = −0.395, S.E. = 0.165) being significantly negative, coefficients of the other three incentive policies and the chemical abandon rate are all significantly positive. The largest coefficient is that of the chemical abandon rate (β = 5.568, S.E. = 2.179), followed by environmental propaganda (β = 0.885, S.E. = 0.185), agricultural insurance (β = 0.470, S.E. = 0.163), and green subsidy (β = 0.273, S.E. = 0.049). It is noteworthy that in this group, the coefficient of ASC (β = 2.876, S.E. = 0.523) is significantly positive (although negative in the other two groups) and considerably large. In this group, it is interesting that farmers are largely self-promoted and only in marginal need of any incentive policies, or even avoid the need for external technical support; thus, we name it the autonomy-orientated group. Among the three groups, this group, with a medium-aged (M_age = 58) and middle-school-educated household head, has the smaller family size (M_family = 5) and the smallest labor population (M_labor = 3); however, its planting area (M_planting = 6.31 hm²) is the largest among all the groups.

5. Discussion

Consistent with previous research, the current study confirms that ecological fertilization and deinsectization can be promoted among small-scale farmers through various policy incentives, including green subsidy [43,44], agricultural insurance [49], green technique support [47], and environmental protection propaganda [48]. However, by introducing farmers’ heterogeneity of characteristics, the present study offers an updated overview of farmers’ responses to the policy incentives, as elaborated in the following subsections.

5.1. Farmers’ Characteristics Influencing Their Policy Incentive Preferences

Among individual-level characteristics, farmers with higher education tend to be more accepting of ecological fertilization and deinsectization (similar to the previous research of [53]). Among household-level characteristics, increasing family size obstructs the household’s adoption of ecological fertilization and pest extermination; this finding is similar to that of previous research such as [52,57]. As for environmental characteristics, farmers who do not participate in agricultural organizations prefer technical support policies compared with farmers who join agricultural organizations (more details in [64]). However, contrary to previous research [53,54], this current study finds that senior farmers seems to be more easily encouraged to adopt ecological fertilization and pest extermination via the provision of green subsidy than their younger peers. This study offers the following explanation—that price constraints restrict senior farmers (who are more economically sensitive [79,80]) from adopting ecological farming; thus, the provision of a government subsidy should help them overcome such price constraints and consequently encourage them to farm ecologically.

5.2. Alternating and Complementing Effects of Incentive Policies

This study finds that, despite technical support and environmental propaganda substituting each other, there are no significant substitution effects among other incentives. The explanation might be due to differences in farmers’ green-agriculture qualifications, endowments, and motivations: farmers inclined toward environmentally friendly livelihoods want to perceive themselves as having an innately responsible attitude toward nature and human society, rather than being facilitated by outsiders’ help [81]; they are also more capable of acquiring green agriculture technology through their social networks [82,83]; thus, they find the afforded-by-government technical support less effective. Moreover, complementary effects exist between agricultural insurance and green subsidy (environmental propaganda). The main reason might be because agricultural insurance, as a post hoc compensation measure, can afford security (either economically or psychologically) when farmers encounter green subsidies or environmental propaganda in the context of the economic risk or psychological vulnerability of agricultural work.

5.3. A Typology of Green Agriculture Incentives

Generally, this study finds that small-scale farmers have three orientations: towards the economy, security, and autonomy. Economy and security orientations dominate farmers’ incentive preference. However, there is a noteworthy proportion (25.0%) of farmers who are intrinsically motivated by their environmental awareness. Fostering this innate motivation can reduce the pressure on governments to invest in green agriculture incentives. More importantly, farmers’ preferences in regard to green agriculture incentive mixes are structured by their circumstances. For instance, younger and well-educated household heads in lightly populated families tend to prioritize economy trade-offs for green agriculture; while senior and lower-educated household heads with a large family who have to survive on a relatively smaller farm have to consider security concerns as a priority; interestingly, despite the household head being older with a low level of education, a lightly populated family with a diversified livelihood of “agriculture + migratory labor” prefers to intrinsically herald in a green agriculture transformation by relying on their own technical resources. Figure 3 provides a profile of small-scale farmers in regard to decision making on various incentive mixes.

6. Implications

6.1. Theoretical Implications

First, by incorporating the interactions of farmers’ characteristics and the policy incentives into the research model, this study reveals relatively comprehensive relationships between the influencing factors of farmers’ policy incentive preference, complements previous (yet fragmented) research on those influencing factors’ effects, and differentiates farmers’ three main motivations: security, economic rewards, and autonomy. Thus, the present study forms a foundation for future research to build a framework of farmers’ incentives and influencing characteristics in the context of adoption of green agriculture.

Second, this current study also emphasizes the interactions between different incentive policies in its research scope, thus enriching the green incentive policy research that previously focused on only a specific single policy. According to this study’s results, there are substitution effects (i.e., between technical support and environmental propaganda) and complementary effects (i.e., between agricultural insurance and green subsidy/environmental propaganda). Thus, the current study forms the basis for more research on the in-depth influencing mechanisms of green incentive policies.

Additionally, the present study confirms that the choice experiment is appropriate for investigating the synthesis of multiple factors and the responses to multiple incentives. Thus, future research on incentive responses could also apply this method.

6.2. Practical Implications

Practically, this study advocates more targeted and coordinated green agriculture incentive policies. First, the incentive policies should fully consider farmers’ heterogeneous characteristics rather than regarding them as a single homogeneous target. Therefore, future policy design, implementation, and evaluation should involve collecting, analyzing, and even transforming those characteristics. For instance, literacy projects such as upgrading farmers’ education levels can be integrated into incentive plans of green farming. Second, the diverse incentive policies should engage in a mutually influencing system instead of contradicting and restricting one another. This requires the government to set up long-term and overall incentive plans instead of implementing specific incentive policies temporarily. Third, the farmer characteristic-incentive typology (refer to Figure 3) could help the local government better target different types of farmers when implementing incentive policies. Generally, China has the world’s most ultra-small-scale farmers and the largest farming area, and is making tremendous efforts in developing green agriculture to protect and make the best use of its farmland. Thus, it is expected that China’s lessons will offer a valuable action reference for other countries (especially for developing countries with a similar planting pattern, e.g., Thailand [84], India [85], Indonesia [86], and the Philippines [87]) practicing green agriculture.

7. Conclusions, Limitations, and Future Directions

7.1. Conclusions

In this paper, a choice experiment was conducted on 1032 grain farmers in Hubei Province. Then, MLM and the LCM were used to study farmers’ preference and heterogeneity of characteristics in regard to the incentive policies of chemical reduction, with the main finding of heterogeneity of farmers’ preferences for relevant incentives. Accordingly, farmers can be divided into economy-orientated, security-orientated, and autonomy-orientated groups: each group prefers a different incentive mix.

7.2. Limitations and Future Directions

This study undeniably has some limitations that can serve as inspiration for future research. For instance, this study only examines four different green agriculture incentive policies. Thus, future research could continue to include more incentives, such as social media influencers [88] and omnichannel integration [89], in the examined model. More importantly, despite the fact that this study explores several characteristic variables of farmers, there is still vast scope for research on emerging characteristics, such as farming households with female versus male heads, farmers’ mobility, the inter-generation wealth status, their internet/social media usage, and their political and pro-social activity participation, and the conflict between scientific knowledge and indigenous traditional knowledge. In addition, farmers’ green agriculture decision models should also include challenges from environmental/economical/societal uncertainties (e.g., the COVID-19 pandemic and natural calamities). Further, other methodologies, such as comparative configuration analysis, could be applied to the interactions among multiple variables. Additionally, considering that this current study only uses a regional sample and cross-sectional data, future research could involve analysis of nation-wide samples, cross-national comparisons, or panel data.